Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x10a4914a8>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x10e022630>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.3.0

/Users/manopaul/anaconda/envs/deep-learning/lib/python3.6/site-packages/ipykernel_launcher.py:14: UserWarning: No GPU found. Please use a GPU to train your neural network.

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    inputs_real = tf.placeholder(tf.float32, shape=(None, image_width, image_height, image_channels), name='input_real')
    inputs_z = tf.placeholder(tf.float32, shape=(None, z_dim), name='input_z')
    lr = tf.placeholder(tf.float32, shape=(), name='learning_rate')
    return inputs_real, inputs_z, lr 

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    alpha = 0.01
    with tf.variable_scope('discriminator', reuse=reuse):
        # 28x28x3 - starting image dimensions
        # Using Xavier_Initializer as per project reviewer's feedback to break symmetry and make model converge faster
        d_layer1 = tf.layers.conv2d(images, 64,5, strides=2, padding='SAME',
                                   kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True,
                                                                                          seed=None,
                                                                                          dtype=tf.float32
                                                                                          ))
        relu1 = tf.maximum(alpha * d_layer1, d_layer1)
        # 14x14x64
        
        d_layer2 = tf.layers.conv2d(d_layer1, 128, 5, strides=2, padding='SAME')
        bn2 = tf.layers.batch_normalization(d_layer2, training=True)
        relu2 = tf.maximum(alpha * bn2, bn2)
        # 7x7x128
        
        d_layer3 = tf.layers.conv2d(d_layer2, 256, 5, strides=2, padding='SAME')
        bn3 = tf.layers.batch_normalization(d_layer3, training=True)
        relu3 = tf.maximum(alpha * bn3, bn3)
        # 4x4x256
        
        # Flatten it
        flat = tf.reshape(relu3, (-1, 4*4*256))
        logits = tf.layers.dense(flat, 1)
        out = tf.sigmoid(logits)
        
    return out, logits

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    
    alpha = 0.01
    with tf.variable_scope('generator', reuse=not is_train):
        # First fully connected layer
        g_layer1 = tf.layers.dense(z, 7*7*256)
        # Reshape it to start the convolutional stack
        g_layer1 = tf.reshape(g_layer1, (-1, 7,7,256))
        g_layer1 = tf.layers.batch_normalization(g_layer1, training=is_train)
        g_layer1 = tf.maximum(alpha * g_layer1, g_layer1)
        # 7x7x256
        
        g_layer2 = tf.layers.conv2d_transpose(g_layer1, 128, 5, strides=2, padding='SAME', 
                                              kernel_initializer=tf.contrib.layers.xavier_initializer(uniform=True,
                                                                                          seed=None,
                                                                                          dtype=tf.float32
                                                                                          ))
        g_layer2 = tf.layers.batch_normalization(g_layer2, training=is_train)
        g_layer2 = tf.maximum(alpha * g_layer2, g_layer2)
        # 14x14x128
        
        g_layer3 = tf.layers.conv2d_transpose(g_layer2, 64, 5, strides=2, padding='SAME')
        g_layer3 = tf.layers.batch_normalization(g_layer3, training=is_train)
        g_layer3 = tf.maximum(alpha * g_layer3, g_layer3)
        # 28x28x64
        
        # Output layer
        logits = tf.layers.conv2d_transpose(g_layer3, out_channel_dim, 5, strides=1, padding='SAME')
        
        out = tf.tanh(logits)
        
    return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)
    
    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, labels=tf.ones_like(d_model_real)))
    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.zeros_like(d_model_fake)))
    
    g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))
    d_loss = d_loss_real + d_loss_fake
    
    return d_loss, g_loss

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # Get weights and bias to update
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]
    
    # Optimize
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)

    return d_train_opt, g_train_opt

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    
    data_image_mode="RGB"
    image_width = data_shape[1]
    image_height = data_shape[2]
    image_depth = data_shape[3]
    
    input_real, input_z, lr = model_inputs(image_width, image_height, image_depth, z_dim)
    d_loss, g_loss = model_loss(input_real, input_z, image_depth)
    d_opt, g_opt = model_opt(d_loss, g_loss, learning_rate, beta1)
    
    saver = tf.train.Saver()
    
    samples, losses = [], []
    steps = 0
    print_every = 10
    show_every = 100
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            for batch_images in get_batches(batch_size):
                batch_images *= 2
                steps += 1
                
                # Sample random noise for Generator
                batch_z = np.random.uniform(-1,1,size=(batch_size, z_dim))

                # Run optimizers
                _ = sess.run(d_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr:learning_rate})
                _ = sess.run(g_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr:learning_rate})
                
                if steps % print_every == 0:
                    # Print losses at end of each epoch
                    train_loss_d = d_loss.eval({input_z: batch_z, input_real: batch_images})
                    train_loss_g = g_loss.eval({input_z: batch_z})

                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count),
                          "Discriminator Loss: {:.4f}...".format(train_loss_d),
                          "Generator Loss: {:.4f}".format(train_loss_g))
                    # Save losses to view after training
                    losses.append((train_loss_d, train_loss_g))

                if steps % show_every == 0:
                    if image_depth == 1:
                            data_image_mode= "L"
                    show_generator_output(sess, 25, input_z, image_depth, data_image_mode)
        
        saver.save(sess, './checkpoints/generator.ckpt')

    return losses

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 16 #128
z_dim = 100
learning_rate = 0.0007
beta1 = 0.2

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/2... Discriminator Loss: 5.5389... Generator Loss: 0.0050
Epoch 1/2... Discriminator Loss: 0.0053... Generator Loss: 18.2402
Epoch 1/2... Discriminator Loss: 0.4529... Generator Loss: 1.0187
Epoch 1/2... Discriminator Loss: 0.0088... Generator Loss: 11.4992
Epoch 1/2... Discriminator Loss: 0.0008... Generator Loss: 8.8445
Epoch 1/2... Discriminator Loss: 0.0229... Generator Loss: 7.9375
Epoch 1/2... Discriminator Loss: 0.0130... Generator Loss: 6.0877
Epoch 1/2... Discriminator Loss: 0.0135... Generator Loss: 10.2666
Epoch 1/2... Discriminator Loss: 0.1101... Generator Loss: 3.0463
Epoch 1/2... Discriminator Loss: 1.6547... Generator Loss: 3.1057

Epoch 1/2... Discriminator Loss: 0.8837... Generator Loss: 1.1439
Epoch 1/2... Discriminator Loss: 1.8704... Generator Loss: 3.7286
Epoch 1/2... Discriminator Loss: 0.4294... Generator Loss: 1.4675
Epoch 1/2... Discriminator Loss: 2.1306... Generator Loss: 0.2455
Epoch 1/2... Discriminator Loss: 1.2016... Generator Loss: 1.0792
Epoch 1/2... Discriminator Loss: 1.1558... Generator Loss: 2.5875
Epoch 1/2... Discriminator Loss: 2.0068... Generator Loss: 0.2337
Epoch 1/2... Discriminator Loss: 1.4287... Generator Loss: 1.4837
Epoch 1/2... Discriminator Loss: 0.8860... Generator Loss: 1.0028
Epoch 1/2... Discriminator Loss: 1.0931... Generator Loss: 1.1276

Epoch 1/2... Discriminator Loss: 0.7919... Generator Loss: 1.0353
Epoch 1/2... Discriminator Loss: 1.4405... Generator Loss: 1.4878
Epoch 1/2... Discriminator Loss: 1.2231... Generator Loss: 0.6028
Epoch 1/2... Discriminator Loss: 1.5101... Generator Loss: 0.4049
Epoch 1/2... Discriminator Loss: 0.5757... Generator Loss: 1.5060
Epoch 1/2... Discriminator Loss: 1.4311... Generator Loss: 0.5073
Epoch 1/2... Discriminator Loss: 1.4060... Generator Loss: 0.5260
Epoch 1/2... Discriminator Loss: 1.1860... Generator Loss: 0.9695
Epoch 1/2... Discriminator Loss: 0.9724... Generator Loss: 1.6295
Epoch 1/2... Discriminator Loss: 0.7166... Generator Loss: 1.3182

Epoch 1/2... Discriminator Loss: 0.8442... Generator Loss: 1.4248
Epoch 1/2... Discriminator Loss: 1.1751... Generator Loss: 0.5674
Epoch 1/2... Discriminator Loss: 1.1402... Generator Loss: 1.0685
Epoch 1/2... Discriminator Loss: 1.1118... Generator Loss: 0.5944
Epoch 1/2... Discriminator Loss: 0.8627... Generator Loss: 0.9046
Epoch 1/2... Discriminator Loss: 0.6163... Generator Loss: 1.1750
Epoch 1/2... Discriminator Loss: 0.7427... Generator Loss: 1.2468
Epoch 1/2... Discriminator Loss: 0.3156... Generator Loss: 1.8109
Epoch 1/2... Discriminator Loss: 1.1730... Generator Loss: 0.7663
Epoch 1/2... Discriminator Loss: 1.1285... Generator Loss: 0.7607

Epoch 1/2... Discriminator Loss: 1.3182... Generator Loss: 1.6695
Epoch 1/2... Discriminator Loss: 0.5718... Generator Loss: 3.5163
Epoch 1/2... Discriminator Loss: 0.6861... Generator Loss: 1.3908
Epoch 1/2... Discriminator Loss: 1.3848... Generator Loss: 0.8102
Epoch 1/2... Discriminator Loss: 1.0405... Generator Loss: 0.6887
Epoch 1/2... Discriminator Loss: 0.7373... Generator Loss: 1.0553
Epoch 1/2... Discriminator Loss: 0.8033... Generator Loss: 1.0434
Epoch 1/2... Discriminator Loss: 0.9273... Generator Loss: 1.0104
Epoch 1/2... Discriminator Loss: 1.0600... Generator Loss: 0.9506
Epoch 1/2... Discriminator Loss: 1.1892... Generator Loss: 0.6743

Epoch 1/2... Discriminator Loss: 1.5196... Generator Loss: 0.3431
Epoch 1/2... Discriminator Loss: 1.2578... Generator Loss: 0.5336
Epoch 1/2... Discriminator Loss: 1.2854... Generator Loss: 0.9542
Epoch 1/2... Discriminator Loss: 1.1677... Generator Loss: 0.7850
Epoch 1/2... Discriminator Loss: 1.4443... Generator Loss: 0.3496
Epoch 1/2... Discriminator Loss: 1.5232... Generator Loss: 0.3617
Epoch 1/2... Discriminator Loss: 1.3911... Generator Loss: 0.5247
Epoch 1/2... Discriminator Loss: 1.4093... Generator Loss: 1.0273
Epoch 1/2... Discriminator Loss: 1.3137... Generator Loss: 0.8657
Epoch 1/2... Discriminator Loss: 1.6281... Generator Loss: 0.3188

Epoch 1/2... Discriminator Loss: 1.4284... Generator Loss: 0.5458
Epoch 1/2... Discriminator Loss: 1.4792... Generator Loss: 0.3791
Epoch 1/2... Discriminator Loss: 1.2636... Generator Loss: 0.5539
Epoch 1/2... Discriminator Loss: 1.2639... Generator Loss: 0.5473
Epoch 1/2... Discriminator Loss: 1.3434... Generator Loss: 0.7846
Epoch 1/2... Discriminator Loss: 1.2200... Generator Loss: 0.6634
Epoch 1/2... Discriminator Loss: 1.3403... Generator Loss: 0.9073
Epoch 1/2... Discriminator Loss: 1.5719... Generator Loss: 0.4359
Epoch 1/2... Discriminator Loss: 1.3885... Generator Loss: 0.3853
Epoch 1/2... Discriminator Loss: 1.3959... Generator Loss: 0.9921

Epoch 1/2... Discriminator Loss: 1.2929... Generator Loss: 0.7058
Epoch 1/2... Discriminator Loss: 1.2866... Generator Loss: 0.7850
Epoch 1/2... Discriminator Loss: 1.2225... Generator Loss: 0.5631
Epoch 1/2... Discriminator Loss: 1.2411... Generator Loss: 0.5707
Epoch 1/2... Discriminator Loss: 1.1444... Generator Loss: 0.6377
Epoch 1/2... Discriminator Loss: 1.1946... Generator Loss: 0.9449
Epoch 1/2... Discriminator Loss: 1.2710... Generator Loss: 0.5938
Epoch 1/2... Discriminator Loss: 1.3287... Generator Loss: 0.6703
Epoch 1/2... Discriminator Loss: 1.4905... Generator Loss: 0.4416
Epoch 1/2... Discriminator Loss: 1.2427... Generator Loss: 0.7019

Epoch 1/2... Discriminator Loss: 1.3075... Generator Loss: 0.5688
Epoch 1/2... Discriminator Loss: 1.3968... Generator Loss: 0.6447
Epoch 1/2... Discriminator Loss: 1.3008... Generator Loss: 0.5287
Epoch 1/2... Discriminator Loss: 1.4225... Generator Loss: 0.5263
Epoch 1/2... Discriminator Loss: 1.2528... Generator Loss: 0.8759
Epoch 1/2... Discriminator Loss: 1.2593... Generator Loss: 0.6644
Epoch 1/2... Discriminator Loss: 1.4722... Generator Loss: 0.3953
Epoch 1/2... Discriminator Loss: 1.1274... Generator Loss: 0.6094
Epoch 1/2... Discriminator Loss: 1.1568... Generator Loss: 1.1037
Epoch 1/2... Discriminator Loss: 1.5150... Generator Loss: 0.3869

Epoch 1/2... Discriminator Loss: 1.7178... Generator Loss: 0.3036
Epoch 1/2... Discriminator Loss: 1.4656... Generator Loss: 0.4533
Epoch 1/2... Discriminator Loss: 1.3918... Generator Loss: 0.5606
Epoch 1/2... Discriminator Loss: 1.2758... Generator Loss: 0.5907
Epoch 1/2... Discriminator Loss: 1.2461... Generator Loss: 0.6929
Epoch 1/2... Discriminator Loss: 1.3940... Generator Loss: 0.4992
Epoch 1/2... Discriminator Loss: 1.3135... Generator Loss: 0.6115
Epoch 1/2... Discriminator Loss: 1.4955... Generator Loss: 0.4284
Epoch 1/2... Discriminator Loss: 1.3825... Generator Loss: 0.5183
Epoch 1/2... Discriminator Loss: 1.5659... Generator Loss: 0.3219

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 16 #128
z_dim = 100
learning_rate = 0.0007
beta1 = 0.2

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/1... Discriminator Loss: 0.0260... Generator Loss: 4.1751
Epoch 1/1... Discriminator Loss: 0.0075... Generator Loss: 11.6688
Epoch 1/1... Discriminator Loss: 2.1732... Generator Loss: 0.1224
Epoch 1/1... Discriminator Loss: 0.0133... Generator Loss: 5.0962
Epoch 1/1... Discriminator Loss: 0.0024... Generator Loss: 14.3046
Epoch 1/1... Discriminator Loss: 0.0147... Generator Loss: 15.2927
Epoch 1/1... Discriminator Loss: 0.0008... Generator Loss: 10.2871
Epoch 1/1... Discriminator Loss: 0.0944... Generator Loss: 2.6973
Epoch 1/1... Discriminator Loss: 0.6655... Generator Loss: 5.8094
Epoch 1/1... Discriminator Loss: 1.8001... Generator Loss: 0.3863

Epoch 1/1... Discriminator Loss: 1.0586... Generator Loss: 0.6627
Epoch 1/1... Discriminator Loss: 0.4479... Generator Loss: 1.6697
Epoch 1/1... Discriminator Loss: 1.5617... Generator Loss: 0.4463
Epoch 1/1... Discriminator Loss: 0.7568... Generator Loss: 1.3717
Epoch 1/1... Discriminator Loss: 0.1183... Generator Loss: 4.4926
Epoch 1/1... Discriminator Loss: 1.1318... Generator Loss: 1.2889
Epoch 1/1... Discriminator Loss: 1.2652... Generator Loss: 0.4300
Epoch 1/1... Discriminator Loss: 0.5540... Generator Loss: 1.2387
Epoch 1/1... Discriminator Loss: 0.4583... Generator Loss: 3.0786
Epoch 1/1... Discriminator Loss: 0.1381... Generator Loss: 2.8490

Epoch 1/1... Discriminator Loss: 0.5074... Generator Loss: 1.1991
Epoch 1/1... Discriminator Loss: 0.2217... Generator Loss: 2.4094
Epoch 1/1... Discriminator Loss: 0.0717... Generator Loss: 3.9979
Epoch 1/1... Discriminator Loss: 0.1052... Generator Loss: 3.2318
Epoch 1/1... Discriminator Loss: 0.0245... Generator Loss: 5.9797
Epoch 1/1... Discriminator Loss: 0.0579... Generator Loss: 4.0027
Epoch 1/1... Discriminator Loss: 0.0396... Generator Loss: 4.6143
Epoch 1/1... Discriminator Loss: 0.0311... Generator Loss: 4.8246
Epoch 1/1... Discriminator Loss: 0.0130... Generator Loss: 5.6197
Epoch 1/1... Discriminator Loss: 0.0142... Generator Loss: 5.2689

Epoch 1/1... Discriminator Loss: 1.5350... Generator Loss: 0.4785
Epoch 1/1... Discriminator Loss: 0.1623... Generator Loss: 3.2646
Epoch 1/1... Discriminator Loss: 0.1006... Generator Loss: 3.3730
Epoch 1/1... Discriminator Loss: 0.0257... Generator Loss: 6.1010
Epoch 1/1... Discriminator Loss: 0.9748... Generator Loss: 6.2483
Epoch 1/1... Discriminator Loss: 0.0267... Generator Loss: 6.9470
Epoch 1/1... Discriminator Loss: 0.0935... Generator Loss: 3.0309
Epoch 1/1... Discriminator Loss: 2.0338... Generator Loss: 4.5775
Epoch 1/1... Discriminator Loss: 0.0567... Generator Loss: 3.8276
Epoch 1/1... Discriminator Loss: 0.0832... Generator Loss: 3.6796

Epoch 1/1... Discriminator Loss: 0.0531... Generator Loss: 3.7602
Epoch 1/1... Discriminator Loss: 0.0554... Generator Loss: 3.8071
Epoch 1/1... Discriminator Loss: 1.1367... Generator Loss: 1.2828
Epoch 1/1... Discriminator Loss: 0.2611... Generator Loss: 3.0258
Epoch 1/1... Discriminator Loss: 0.0723... Generator Loss: 3.1010
Epoch 1/1... Discriminator Loss: 3.4551... Generator Loss: 0.0667
Epoch 1/1... Discriminator Loss: 2.6479... Generator Loss: 0.0870
Epoch 1/1... Discriminator Loss: 0.3378... Generator Loss: 2.8158
Epoch 1/1... Discriminator Loss: 0.4034... Generator Loss: 2.5218
Epoch 1/1... Discriminator Loss: 1.6997... Generator Loss: 0.4234

Epoch 1/1... Discriminator Loss: 1.3708... Generator Loss: 0.3637
Epoch 1/1... Discriminator Loss: 1.8378... Generator Loss: 6.2570
Epoch 1/1... Discriminator Loss: 0.6451... Generator Loss: 0.9845
Epoch 1/1... Discriminator Loss: 0.1167... Generator Loss: 3.1307
Epoch 1/1... Discriminator Loss: 0.3838... Generator Loss: 1.4234
Epoch 1/1... Discriminator Loss: 0.2889... Generator Loss: 1.7419
Epoch 1/1... Discriminator Loss: 0.1381... Generator Loss: 2.7326
Epoch 1/1... Discriminator Loss: 0.6089... Generator Loss: 0.9346
Epoch 1/1... Discriminator Loss: 0.9904... Generator Loss: 2.0782
Epoch 1/1... Discriminator Loss: 0.2523... Generator Loss: 3.6251

Epoch 1/1... Discriminator Loss: 0.1908... Generator Loss: 3.4802
Epoch 1/1... Discriminator Loss: 1.0841... Generator Loss: 0.6114
Epoch 1/1... Discriminator Loss: 0.3783... Generator Loss: 1.3780
Epoch 1/1... Discriminator Loss: 0.9235... Generator Loss: 1.7510
Epoch 1/1... Discriminator Loss: 0.2280... Generator Loss: 2.3086
Epoch 1/1... Discriminator Loss: 1.1229... Generator Loss: 2.4034
Epoch 1/1... Discriminator Loss: 0.4403... Generator Loss: 2.2643
Epoch 1/1... Discriminator Loss: 0.0799... Generator Loss: 3.9842
Epoch 1/1... Discriminator Loss: 1.4652... Generator Loss: 3.7732
Epoch 1/1... Discriminator Loss: 1.3096... Generator Loss: 2.7748

Epoch 1/1... Discriminator Loss: 1.0486... Generator Loss: 2.7200
Epoch 1/1... Discriminator Loss: 1.2013... Generator Loss: 0.5098
Epoch 1/1... Discriminator Loss: 1.5550... Generator Loss: 3.0271
Epoch 1/1... Discriminator Loss: 1.0872... Generator Loss: 0.5691
Epoch 1/1... Discriminator Loss: 0.3793... Generator Loss: 1.5506
Epoch 1/1... Discriminator Loss: 1.8741... Generator Loss: 3.4592
Epoch 1/1... Discriminator Loss: 0.3708... Generator Loss: 2.0723
Epoch 1/1... Discriminator Loss: 0.2506... Generator Loss: 2.4876
Epoch 1/1... Discriminator Loss: 1.3148... Generator Loss: 2.2487
Epoch 1/1... Discriminator Loss: 0.4937... Generator Loss: 2.3872

Epoch 1/1... Discriminator Loss: 0.6546... Generator Loss: 0.9979
Epoch 1/1... Discriminator Loss: 0.8984... Generator Loss: 0.9188
Epoch 1/1... Discriminator Loss: 1.2869... Generator Loss: 1.6564
Epoch 1/1... Discriminator Loss: 0.7382... Generator Loss: 1.8424
Epoch 1/1... Discriminator Loss: 0.3888... Generator Loss: 3.7778
Epoch 1/1... Discriminator Loss: 0.3800... Generator Loss: 1.5353
Epoch 1/1... Discriminator Loss: 2.9266... Generator Loss: 3.3126
Epoch 1/1... Discriminator Loss: 1.2404... Generator Loss: 3.4955
Epoch 1/1... Discriminator Loss: 1.2417... Generator Loss: 0.5089
Epoch 1/1... Discriminator Loss: 0.8170... Generator Loss: 0.7686

Epoch 1/1... Discriminator Loss: 0.8755... Generator Loss: 0.7844
Epoch 1/1... Discriminator Loss: 0.4709... Generator Loss: 1.7988
Epoch 1/1... Discriminator Loss: 0.7918... Generator Loss: 1.0788
Epoch 1/1... Discriminator Loss: 1.8166... Generator Loss: 0.2047
Epoch 1/1... Discriminator Loss: 0.4649... Generator Loss: 1.4145
Epoch 1/1... Discriminator Loss: 0.7566... Generator Loss: 0.9084
Epoch 1/1... Discriminator Loss: 0.4321... Generator Loss: 1.8866
Epoch 1/1... Discriminator Loss: 1.8158... Generator Loss: 4.7090
Epoch 1/1... Discriminator Loss: 0.6488... Generator Loss: 0.9402
Epoch 1/1... Discriminator Loss: 1.0905... Generator Loss: 2.5841

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.